The numbers Brief Episode map Takeaways Charts Quotes Clips Glossary Go next Transcript

Episode 150 · Fintech · 84 min

The country that trades before it invests

Systematic strategies are under 1% of India's fund market — the share the US last had in 1986. Keith's claim is that the derivatives crowd everyone calls reckless is the on-ramp, not the problem: something like 80% of the world's F&O participants by headcount are Indian, and they will age into owning equities. InvestorAi's answer is a computer-vision model that turns each trading day into 32x32-pixel images and reads about 1.5 trillion data points.

AI & machine learning Savings & wealth India macro

Bruce Keith

Co-founder & CEO, InvestorAi · with Vishal Krishna

1:23:55

Said in this episode

▶ 3:26

<1%

Systematic strategies' share of India's fund market

Against 60-plus per cent active and 40-odd per cent passive; Keith says the US last had that mix in 1986 and is now roughly 25% active, 35% quant, the rest passive.

▶ 4:36

500x

Keith's headroom for quant investing in India

15x of household mutual fund penetration multiplied by 35x of quant market share, assuming no growth in the market at all — arithmetic he does live, using mutual funds as a proxy.

▶ 6:57

80%

India's share of world F&O participants

Counted by number of people in the derivatives market rather than value traded — the figure Keith uses to argue Indians have a trading mindset closer to America's than Europe's.

▶ 16:47

1.5 trillion

Data points the model reads

1,024 features per 32x32 image, times the Nifty 500, times about 250 trading days, times twelve years — built on a calculator during the interview.

▶ 23:56

2.1 billion

People with enough liquid capital to be advised

Out of a planet Keith puts at 8.6 billion; he says only a small fraction are actually advised, leaving roughly 2.05 billion unserved — the captions garble the advised figure, so treat the sliver as directional.

▶ 53:34

15% → 70%

Share of InvestorAi's code written by machines

Where it stands now versus where he hopes to be by the end of the year — the reason he says he does not need 500 engineers, only people who can judge the output.

The brief

The argument in sixty seconds

Keith's claim is that India's investing market is mis-read by almost everyone watching it. Using mutual funds as a proxy, systematic and quant strategies are under 1% of the Indian industry — active is 60-plus per cent, passive 40-odd — a mix the US last had in 1986, against today's American split of roughly 25% active, 35% quant and the rest passive. Layer 15x of household fund penetration on 35x of quant share and he gets a 500x opportunity without assuming any market growth at all. The bridge, he argues, is the thing India's commentariat calls a problem: the country holds something like 80% of the world's derivative participants by headcount, Dream11's active subscribers still outnumber mutual fund investors by a multiple, and those young traders are learning about money before they age into owning equities the American way rather than the European way. InvestorAi's method is deliberately not old-school quant, whose edges arbitrage away in weeks — financial data is converted into 32x32-pixel images and read by computer vision across Nifty 500 stocks and twelve years, about 1.5 trillion data points, with a second layer of AI choosing which model runs in which market. Because the neural net cannot explain itself, every rebalance is percentile-scored against the best twelve stocks it could have picked. He is blunt about the limits — nobody guarantees returns, the product is an Iron Man suit and not a Superman one — and about where the real risk sits: not in the machine, but in a generation of AI engineers building on other people's platforms with no structure underneath them.

Worth your time if you are

Retail investors who found the trading app before the strategy

Brokers and wealth managers deciding whether to build AI or rent it

AI engineers who have only ever built on someone else's platform

Anyone underwriting the India growth story for the next twenty years

Episode map

Where the conversation travels

Every block is a chapter, coloured by what it's about. Click any of it to jump straight to that minute on YouTube.

AI & machine learningSavings & wealthIndia macroFounder journeyHiring & talentRegulation & policy

01Cold open: the arbitrage that vanished 0:00 Keith first came to India in 1991 and started working here in the offshoring years of 2005, and says the wage arbitrage that brought him has gone — at senior levels he now pays the same in Bangalore as he would in Europe — leaving the domestic market and twenty-plus years of demographic growth as the only reason to be here. 02Quant is under 1%, with 35x to run 2:21 Taking the mutual fund industry as a proxy, Keith lays out the gap — India is 60-plus per cent actively managed, 40-odd passive and under 1% systematic, the mix the US last had in 1986, against America's present 25% active and 35% quant — then multiplies 15x of household penetration by 35x of quant share into a 500x opportunity before any market growth. 0380% of the world's F&O traders 5:56 Asked whether investing will be Indianised the way McDonald's was, Keith answers with scale — India holds something like 80% of the world's derivative participants by number of people, and Dream11's active subscribers still outnumber mutual fund investors by a multiple — then refuses to police Gen Z, backs SEBI's lot-size changes and defends a regulator he thinks is in a no-win position. 04Old-school quant meets the neural net 11:12 The rules, maths and mean-reversion AI of the early 2000s still works, but in the US a clever quant edge is arbitraged away almost as soon as it is found, whereas a learning model compounds like Scotch — and a machine trained on COVID data reads the current market wobble better because it has seen something like it before. 05A calculator, 1.5 trillion, and the audit 15:15 A calculator is fetched on air to build the number — 32x32-pixel images of financial data, 1,024 features a stock, Nifty 500 names, 250 trading days, twelve years — before Keith explains why choosing and ordering those features changes the prediction, why AI sits on top of AI to pick the right model for the market, and how every rebalance is percentile-scored against the best twelve stocks the universe could have offered. 06Two billion people nobody advises 22:25 Everyone tells you when to buy and nobody tells you when to sell, which is Keith's case for continual systematic churn — and for a market he sizes at 2.1 billion people with enough liquid capital to be advised, of whom only a sliver actually are. 07Liberation day as the zenith 25:28 Money migrates and then hides — Switzerland at every moment of risk — and Keith calls what was proudly trumpeted as liberation day the moment we will look back on in twenty years as the zenith of US power, an act of self-harm bigger and longer-lasting than Brexit that accelerates South and North Asia. 08Tooting, returning talent, and Bangalore 28:55 The diaspora arc runs from a 1980s flight with three Indians on it to a Heathrow-Las Vegas flight full of them, and from a senior search aimed at Indian talent in an unwelcoming US that ended in Singapore, to the reason InvestorAi sits here at all — a business plan written in a Bangalore hotel business centre in 2011, built out with the overwhelming majority of its people in India, and sold to State Street. 09Energy is the chink in the armour 34:20 Against the host's Chomsky-sourced claim that finance has displaced manufacturing in American corporate profits, Keith names energy dependence as India's one structural weakness, assumes it is cracked by around 2030, and argues India can grow the way the US did in the 1950s — on self-created domestic demand rather than exports. 10Not manufacturing versus services — IP 37:50 Keith reframes the national argument he says is slightly wrong: the deficit is not factories but intellectual property, visible in AI CVs where mid-level engineers have done a little LLM work on someone else's platform, and in graduates parked in bank and asset-manager AI teams without the structured growth an Infosys or TCS once provided. 115x revenue, five to ten hires 41:10 Agentic AI is defined as an LLM context layer plus orchestration and then immediately costed — an 85%-automated market data process whose exceptions a machine can now investigate and justify, and a 5x revenue target at a just-under-50-person firm that expects to add only five to ten people. 12Twenty brokers and a Dubai fund 45:40 Agentic improves the surface layer rather than the core models, and distribution runs through just shy of twenty broker channel partners on revenue share — app-discovered intraday products at one end, relationship-sold subscription baskets at the other, with a US long-short strategy for a Dubai fund close to going live and the cost of research falling towards nothing. 13Publish the standard before the regulator 50:00 With no audit standard for investment AI in India, Keith is taking his company offsite a challenge — write and publish the framework for judging systematic models instead of waiting for the regulator, and keep publishing until it becomes or shapes the market standard — while conceding everyone has been drinking the AI Kool-Aid and that 15% of his own code is already machine-written. 14The Iron Man suit, and his own book 53:50 India's brokers solved access through eKYC and UPI, so the next contest is research — where Keith argues machines should coexist with analysts because AI is brilliant across 1.5 trillion data points and humans decide well on almost none — before opening his own portfolio: half passive index exposure, a little under a third systematic, the rest yield, commodities and stocks he holds. 15Minefields and a million pieces of glass 1:00:20 The pace-of-change riff runs from carbon copies and a physical in-box to cheap data and mobile-first defaults, then darkens into drone warfare and the self-repositioning minefield an AI godfather described, before landing on Keith's own news diet — BBC front pages and a printed paper on the way to the office — and the conclusion that trust has fractured into a million pieces of glass. 16Nine pence, and a cheque every time 1:11:00 Keith's father was an orphan who left school without a leaving certificate, begged a free bus ride home from a job that evaporated on day one, and built the largest glazing business in the north-east of Scotland — paying his 11-year-old son 9p an invoice, 11p if a cheque came back, which is how a boy did credit control before he knew the phrase; the episode closes on living by outputs, 150 books on an iPad and Birdsong.

Takeaways

Ideas to carry out of this hour

India's fund market is where America stood in 1986

On Keith's numbers, actively managed funds are 60-plus per cent of the Indian mutual fund industry, passive 40-odd, and systematic or quant strategies under 1% of market share. The US last looked like that in 1986; today it is roughly 25% active, 35% quant and the balance passive. That is about 35x of share for quant to travel on a static market — and household ownership of funds has another 15x to go before India looks like the West. Multiply the two and he gets 500x, all of it before any growth in the market itself.

▶ 3:26 ▶ 4:02 ▶ 4:36

The F&O crowd is the on-ramp, not the accident

India holds something like 80% of the world's derivative participants counted by people rather than value, and Dream11's active subscribers still run to a multiple of the country's mutual fund investors. Keith reads that as demand, not pathology: if these young traders were not doing F&O they would be doing crypto, and access via cheap accounts is a good thing his own generation never had. His prediction is that equity ownership in India looks like the US in twenty years rather than like Europe — reached through trading, because trading teaches people about money before they get serious.

▶ 6:57 ▶ 8:09 ▶ 9:10

Old quant gets arbitraged away; a learning model ages

The AI that hedge funds ran on in the early 2000s was logic and reasoning — rules, maths, statistics, reversion to the mean — and in the most advanced market on earth any interesting quant edge disappears as soon as enough money chases it. A model that keeps learning is different: Keith's analogy is Scotch, where a one-year-old will never catch an eight-year-old even as the gap between eight and fifteen narrows. His live example is the current market wobble, which a machine trained on COVID data handles better than one that never saw it, because it recognises the investing pattern rather than the geopolitics behind it.

▶ 12:11 ▶ 13:27 ▶ 14:46

Turn the market into a picture and let a vision model stare

InvestorAi converts financial data into images and runs computer vision over them — the technique used on faces and X-rays. Each image is 32x32, so 1,024 features per stock per day; across the Nifty 500, roughly 250 trading days and twelve years, that is about 1.5 trillion data points, a number the two men build live on a calculator. The hard part is not the volume but the composition: there are thousands of candidate features, only 1,024 fit, and because a vision model reads associations across the grid, changing the order of the columns can gut the predictive power.

▶ 15:45 ▶ 16:47 ▶ 17:46

A black box needs a scoreboard, not a promise

Because it is a neural network it is impossible to know exactly what drove any prediction, so Keith's team works backwards from feature weightings to build a story about each call. Then they measure it: at the end of every two-, four- or six-week rebalance they compute every combination of stocks to find the best twelve the Nifty 500 could have given, and percentile-rank their own picks against it. A drifting percentile is the signal to retrain or change features. The promise made to a customer is correspondingly narrow — a published live track record since 2021 with good and bad periods, and consistency.

▶ 19:48 ▶ 20:34 ▶ 21:34

The gap is advice, and AI makes advice nearly free

By Keith's count 2.1 billion people on the planet hold enough liquid capital to be advised, from robo-advisory up to white-glove Swiss banking, and only a fraction of them actually are. The status quo he is displacing is the WhatsApp group, where everyone tells you when to buy and nobody tells you when to sell. What changes the arithmetic is that AI brings the cost of research down to almost nothing: he can generate literally thousands of distinct portfolios without adding to his cost base, and the machine never takes a holiday or a sick day.

▶ 23:24 ▶ 23:56 ▶ 48:17

Agentic AI is a growth plan, not a headcount plan

InvestorAi's market data pipeline is already about 85% automated, and the residue — stale feeds, spikes, corporate actions, news checks — is exactly the investigative work an agent with context and orchestration can now do, with its reasoning attached for a human to sense-check. The consequence is stated as a target: 5x revenue this year on a base of just under 50 people, adding five to ten, several of them in sales. Agentic will not run the core models, he insists; it improves the surface layer, letting the firm create and analyse product faster.

▶ 42:44 ▶ 44:02 ▶ 45:48

India's shortfall is intellectual property, and the talent path that builds it

The manufacturing-versus-services debate is slightly wrong, Keith argues — India does not create enough intellectual property. He sees it in hiring: every CV now claims AI/ML, but at mid level most people have done a little LLM work on top of somebody else's platform rather than building foundational models, which is why his own senior search ran through the US and Singapore. The structural cause is that graduates who would once have entered a TCS or Infosys programme now join banks and asset managers that put them in a room and say go make it work. With agentic AI about to shrink that demand, his advice is unsentimental: without a real growth path, move.

▶ 37:40 ▶ 38:31 ▶ 40:40

The numbers, drawn

What the episode measures

Every figure below was said on air — timestamps included, caveats kept.

Conversation share

portion of the hour spent on each theme

AI & machine learning24%

Savings & wealth22%

India macro16%

Founder journey13%

Hiring & talent8%

Regulation & policy7%

Computed from the chapter map of this episode.

How fund money is managed: India vs the US

% of market share

As stated in conversation, market share rather than AUM: India '60-plus per cent' active, '40-odd' passive (he hedges with 'or 80/20, whatever') and 'less than 1%' systematic — a split that does not reconcile to exactly 100; the US today 'roughly 25% active, 35% quant and the balance passive'.▶ 3:26

The multipliers Keith is underwriting

x versus today

Keith's own arithmetic on air, using the mutual fund industry as a proxy and assuming no growth in the market itself.▶ 4:36

What the CEO actually owns

% of his portfolio

Passive index ETFs50%

Systematic28%

Yield, commodities, held stocks22%

As described on air: 50% completely passive (Nifty, S&P 500, some FTSE and euro exposure, a little China), systematic 'under a third but more than a quarter' — plotted at 28 — and the remainder in yield, commodities and specific stocks he holds.▶ 58:30

Worth keeping

Lines that stay

Everyone tells you when to buy. No one tells you when to sell.

— Bruce Keith ▶ 23:24

I think what was proudly trumpeted as liberation day — I think we will look back in 20 years and see that as the zenith of US power.

— Bruce Keith ▶ 26:48

If you're recently graduated and you're in a non-tech firm and you don't have a proper career growth path, you're done. I would honestly move.

— Bruce Keith ▶ 40:40

It's an Iron Man costume. You're still flying it, but you've got all these superpowers around you. It's not going to guarantee that you win every time, but it's going to improve your chances.

— Bruce Keith ▶ 59:52

He said, if you come back with a cheque from any one of these people, I'll make it up to 11p. Every single invoice I delivered, I came back with a cheque. I didn't realise I was doing credit control at 11 years old.

— Bruce Keith ▶ 1:15:03

Clips that travel

Short on time? Start here

Wealth managers sizing India's fund market

The 500x arithmetic

The active/passive/quant split, the 1986 comparison with the US, and the 15x times 35x multiplication done live.

2:21 → 5:56 · 4 min ▶ Watch clip

Anyone who wants to know what 'AI investing' actually does

A calculator and 1.5 trillion data points

Computer vision applied to markets: 32x32-pixel images, 1,024 features a stock, and why reordering the columns wrecks the prediction.

15:15 → 18:50 · 4 min ▶ Watch clip

Investors reading geopolitics into portfolios

Liberation day as the zenith

Money's migration, dedollarisation nobody in Europe has heard of, and a self-harm Keith rates bigger than Brexit.

25:28 → 28:55 · 3 min ▶ Watch clip

Operators writing an agentic AI business case

5x revenue, five to ten hires

What agentic actually replaces inside a 50-person firm, and the revenue target attached to it.

41:10 → 45:40 · 4 min ▶ Watch clip

Founders and parents thinking about work ethic

Nine pence, and a cheque every time

The orphan father, the glazing business he built in Aberdeen, and credit control learned at eleven for 2p a cheque.

1:11:00 → 1:18:20 · 7 min ▶ Watch clip

Glossary

The jargon, unpacked

Systematic investing

Using machines to pick a portfolio and then re-pick it on a fixed cycle, rather than making one-off discretionary calls — the continual reinvestment loop Keith says most Indian investors skip.

Quant fund

A fund run on rules, mathematics, statistics and reversion to the mean; Keith calls this the old school of AI, distinct from the neural-network models behind today's systems.

F&O

Futures and options — exchange-traded derivatives; India holds something like 80% of the world's participants by headcount, which is why SEBI has been raising contract lot sizes.

Nifty 500

The index of India's 500 largest listed companies — the universe InvestorAi's models score and the source of its twelve-stock baskets.

Rebalance

The scheduled point — every two, four or six weeks — at which a systematic portfolio's holdings are re-picked, and the moment Keith's team audits the model's percentile performance.

Computer vision

AI that reads images, normally faces or X-rays; InvestorAi converts financial data into 32x32-pixel pictures so a vision model can find patterns across 1,024 features at a time.

Agentic AI

An LLM-driven layer that supplies context and orchestration so software can investigate an exception, decide and explain itself — not merely execute a fixed rule.

MTF

Margin trading facility — broker-funded leverage on share purchases, one of the product lines Keith says is about to launch alongside intraday and rebalanced baskets.

Connections

If this resonated, go here next

EP 58 · Fintech

Full transcript

The whole conversation, searchable

Auto-generated captions, lightly cleaned. Click a timestamp to open that moment on YouTube.